The verb slot in causative constructions. Finding the best ... · Constructions SV1-3/2006 (,...

Constructions SV1-3/2006 (www.constructions-online.de, urn:nbn:de:0009-4-6741, ISSN 1860-2010)

The verb slot in causative constructions. Finding the best fitThe verb slot in causative constructions. Finding the best fitThe verb slot in causative constructions. Finding the best fitThe verb slot in causative constructions. Finding the best fit

Gaëtanelle Gilquin

F.N.R.S. – University of Louvain, Belgium

Abstract Abstract Abstract Abstract Using the technique of multiple distinctive collexeme analysis, this paper seeks to determine the verbs that are

distinctively associated with the non-finite verb slot of English periphrastic causative constructions. Not only does the

analysis reveal that the various causative constructions are attracted to essentially different verbs, but by examining

how these verbs fall into semantic classes, it also hints at subtle differences in meaning between the constructions. In

addition, the paper shows how the technique of multiple distinctive collexeme analysis can be usefully combined with

other, complementary methods, and briefly discusses a number of factors which influence the results of multiple

distinctive collexeme analysis and should therefore ideally be taken into account.

1. Introduction1. Introduction1. Introduction1. Introduction

It is widely accepted, if only implicitly, that periphrastic causative constructions are “always safe”

(Stocker 1990: 61) and that causatives such as cause, get, have or make can therefore be combined

with any verb to express causation. This paper, by contrast, argues that certain verbs are more

likely than others to occur in a particular causative construction. By means of a technique known

as multiple distinctive collexeme analysis, it sets out to bring to light the “collexemes” of the

causative constructions under investigation, i.e. the verbs which are the most distinctive for each

of these constructions. As will be shown, not only does this technique make it possible to

distinguish between several constructions that are regularly considered interchangeable in the

literature, but it also helps assign each of them a semantic value by examining the classes of verbs

attracted to it.

After taking stock of what is known about the verb slot of causative constructions, the data

and methodology used in this study will be briefly presented and two possible objections will be

considered. The results of the multiple distinctive collexeme analysis of causative constructions

will then be discussed and, when necessary, supplemented with findings obtained through other

GAËTANELLE GILQUIN


2

methods. Finally, it will be shown that some stylistic, formal and semantic parameters can have an

influence on the results and hence should ideally be taken into account in a multiple distinctive

collexeme analysis.

While this article is an illustration of the way a multiple distinctive collexeme analysis can

be carried out and how its results can be exploited to describe the particular meanings of several

related constructions, it is also a plea for a more integrated approach to the analysis of collexemes,

one that would integrate additional parameters shown to be influential, but also additional

techniques when their results can fruitfully be combined with the multiple distinctive collexeme

analysis.

2. T2. T2. T2. The verb slot of causative constructions in the literaturehe verb slot of causative constructions in the literaturehe verb slot of causative constructions in the literaturehe verb slot of causative constructions in the literature

Ten different periphrastic causative constructions, illustrated in table 1, have been chosen to form

the basis of this analysis. The focus will be on the verb slot of these constructions (V),1 which can

be realised as an infinitive preceded by to (to-inf), a bare infinitive (inf), a past participle (pp) or a

present participle (prp). In what follows, the causative verb (cause, get, have or make) will be

referred to simply as “causative”, in order to avoid any possible confusion with the verb of the V

slot. In addition, the terms CAUSER, CAUSEE and PATIENT will be used to refer to the different

participants involved in the causative process, cf.2

(1) [The drought] (CAUSER) has causedcausedcausedcaused [millions of people] (CAUSEE) to leave [their

homes] (PATIENT).

(B12 1050)

1 This is not to say, of course, that this is the only slot in causative constructions that is worth investigating. The

nominal element of the non-finite clause, for instance, might also have a role in distinguishing between the various

constructions. Thus, according to Deutschbein and Klitscher (1959: 136), the past participle construction with get is

particularly common with body parts or clothes, e.g. I gotgotgotgot my bootsmy bootsmy bootsmy boots mended. The V slot, however, is central in

determining the type of causation expressed by the construction and is therefore particularly interesting. 2 In this example and the following ones, the causative is in bold and the V slot in italics. The code between brackets

is the reference of the sentence in the British National Corpus.

THE VERB SLOT IN CAUSATIVE CONSTRUCTIONS. FINDING THE BEST FIT


3

ConstructionConstructionConstructionConstruction ExampleExampleExampleExample

[X CAUSE Y Vto-inf] The recession causedcausedcausedcaused the price of aluminium to fall.

(KP0 846)

[X GET Y Vto-inf] Why can’t he getgetgetget Jes to do it?

(KB1 4489)

[X GET Y Vpp] We’ll getgetgetget everything sorted out this week.

(KD3 2801)

[X GET Y Vprp] I couldn’t getgetgetget these earphones working.

(KDH 539)

[X HAVE Y Vinf] Jane hashashashas Roche inspect the hut.

(A05 47)

[X HAVE Y Vpp] Did you havehavehavehave the blades sharpened?

(KCB 674)

[X HAVE Y Vprp] I’ve not hadhadhadhad a blow lamp going this morning.

(KBP 4407)

[X MAKE Y Vinf] It makesmakesmakesmakes our house and garden seem so small.

(KDE 2533)

[X BE made Vto-inf] They’re being taken to court and mademademademade to pay.

(KBF 6320)

[X MAKE Y Vpp] They mademademademade their voices heard at the conference.

(CNA 215)

Table 1. Periphrastic causative constructions

While periphrastic causative constructions have been dealt with a lot in the literature and

from many different perspectives (e.g. Kastovsky 1973; Kemmer & Verhagen 1994; Wierzbicka

1998), there is one aspect that has been largely ignored, namely the question of whether causative

constructions exhibit any lexical preferences with respect to the V slot. In other words, can we

assume that all verbs are equally likely to occur in a causative construction, or do causatives

prefer the company of certain verbs, possibly specific to each construction?

GAËTANELLE GILQUIN


4

Paradoxically, the few scholars who do study the nature of the V slot tend to focus on the

least frequent construction, [X MAKE Y Vpp].3 Van Ek and Robat (1984: 327) restrict this

structure to “collocations denoting the exercise and recognition of influence in the widest sense”.

According to Van Roey (1982: 84), it is especially common with felt and known, a hypothesis

which is nicely illustrated by the following example, taken from Eckersley & Eckersley (1967:

239):

(2) He soon mademademademade his presence felt and his wishes known.

Berland-Delépine (1990: 173) gives the following list of possible past participles: heard,

obeyed, respected and understood, which can all be said to involve some sort of influence. The

same is true of Van Ek’s (1966) corpus-derived list: felt (five occurrences), heard (two

occurrences), respected, established, understood and valued (one occurrence each).

Van Ek (1966) also provides us with some information about two other constructions,

namely [X MAKE Y Vinf] and [X GET Y Vprp].4 The former is said to be common with the

following verbs: feel (12.5% in Van Ek’s data), look and seem (8.75% each), think (5%), laugh

and smile (4.4% each), while the latter arguably tends to contain verbs denoting motion such as

coming or going.5

Francis et al.’s (1996: 306) Collins COBUILD Grammar Patterns – Verbs, finally, contains

quite a long list of examples for [X GET Y Vpp] and [X HAVE Y Vpp] (cf. table 2), but no real

3 This construction has a relative frequency of only 0.32 per 100,000 words in the corpus data, as opposed to e.g.

11.73 for [X MAKE Y Vinf]. 4 He actually discusses all the ten patterns investigated here (although not specifically in their causative use). For most

of them, however, his description of the V slot is stated in very general terms, with indications such as ‘activity’,

‘state’ or ‘occurrence’. 5 It should be emphasised, however, that Van Ek does not distinguish between the causative and the other uses of the

construction, and most of the examples he gives with a verb of motion are not causative, cf. But I suppose you getgetgetget

dozens of women throwing themselves at your head.



5

distinction is made between the two constructions, with only seven patterns presented as solely

characteristic of have (in bold in the table).6

have a limb amputatedhave a limb amputatedhave a limb amputatedhave a limb amputated get/have something overhauled

get/have your teeth capped have someone pagedhave someone pagedhave someone pagedhave someone paged

get/have a job costed get/have your hair permed

get/have your hair cut get/have your ears pierced

get/have your house decorated get/have something printed

get/have your windows double-glazed have your stomach pumpedhave your stomach pumpedhave your stomach pumpedhave your stomach pumped

have a tooth extractedhave a tooth extractedhave a tooth extractedhave a tooth extracted get/have something remade

get/have a prescription filled get/have something repaired

get/have something fixed get/have your house rewired

have someone followedhave someone followedhave someone followedhave someone followed get/have your car serviced

get/have yourself immunized have someone tailedhave someone tailedhave someone tailedhave someone tailed

have a boil lancedhave a boil lancedhave a boil lancedhave a boil lanced get/have yourself vaccinated

get/have something made get/have something valued

get/have something mended get/have your legs waxed

get/have an animal neutered

get/have a job costed out get/have something printed up

get/have your house done up get/have a tooth taken out

get/have a washing machine plumbed in

Table 2. Frequent noun phrases and verbs in [X GET Y Vpp] and [X HAVE Y Vpp] (Francis

et al. 1996: 306)

The V slot of the other periphrastic causative constructions is not dealt with at all in the

literature. This lack of attention for the lexical preferences of causative constructions can probably

be linked to the fact that few linguists have relied on corpus data to investigate these constructions

(some exceptions are Kemmer 2001; Stefanowitsch 2001 or Hollmann 2003, as well as Van Ek

1966 and Francis et al. 1996, cited above, but who do not deal exclusively with causative

6 It should also be added that some of the combinations mentioned in this table would be more appropriately labelled

as experiential, rather than causative constructions (cf. have a limb amputated).

GAËTANELLE GILQUIN


6

constructions). While one can arrive at a relatively good description of the syntactic and (perhaps

more tentatively) semantic behaviour of causatives using one’s intuition alone, it is more difficult

to make reliable claims about their collocational behaviour, that is the preferential lexical

company they keep, for, as Sinclair (1991: 116) points out, collocations are often perceived at a

subliminal level and hence cannot be retrieved by introspection. To remedy this situation, the

present study is based on a large set of authentic examples, from which the V slot is directly

observable. The data and the way they were collected are described in the next section.

3. Collecting the corpus data3. Collecting the corpus data3. Collecting the corpus data3. Collecting the corpus data

Use was made of a subset of the British National Corpus, World Edition (BNC 2000), made up of

five million words of written English (academic prose) and five million words of spoken English

(mainly spontaneous conversations). The collection of the data involved three stages. First,

syntactic patterns corresponding to the ten causative constructions were extracted automatically by

means of BNCweb 2.0 (2002). The search string specified the causative (cause, get, have or

make), the form of the V slot (inf, to-inf, pp or prp) and the maximum span between the two.7

Second, all the hits of the query were examined one by one in order to discard sentences that had

the same structure as causative constructions but were not causative (e.g. a structure with an

infinitive of purpose such as Rules are made to be broken, or a construction such as She had her

car stolen which, according to the most likely interpretation, is experiential).8 Third, the V slot of

each construction was identified manually and its lemma was encoded in a database.9

7 Since the BNC is only tagged (i.e. annotated for part of speech), not parsed (i.e. annotated for syntactic structure),

no attempt was made to specify the nature of the intervening noun phrase. 8 Note that a number of concordances were ambiguous between the causative and non-causative reading. In these

cases, the decisive factor was context. When the context was unclear but the causative interpretation was theoretically

possible, the sentence was included in the data. 9 Spelling variants were grouped together (e.g. realise and realize or co-operate and cooperate). Phrasal verbs, on the

other hand, were treated separately.



7

___________________________________________________________________________

ConstructionConstructionConstructionConstruction SpeechSpeechSpeechSpeech WritingWritingWritingWriting TotalTotalTotalTotal

________________________________________________________________________________________________________________________

[X MAKE Y Vpp] 9 22 31

[X HAVE Y Vprp] 70 3 73

[X HAVE Y Vinf] 48 29 77

[X be made Vto-inf] 23 80 103

[X GET Y Vprp] 130 4 134

[X CAUSE Y Vto-inf] 15 207 222

[X GET Y Vto-inf] 350 52 402

[X HAVE Y Vpp] 637 50 687

[X GET Y Vpp] 809 19 828

[X MAKE Y Vinf] 898 258 1,156

TotalTotalTotalTotal 2,989 724 3,713

________________________________________________________________________________________________________________________

Table 3. Frequency of periphrastic causative constructions

Table 3 shows the frequency of causative constructions in speech, writing and in the whole

corpus.10 It will be noticed that, with the exceptions of [X MAKE Y Vpp], [X BE made Vto-inf] and

[X CAUSE Y Vto-inf], the constructions are more frequent in speech than in writing,11 sometimes

considerably so (cf. [X GET Y Vpp], with 809 occurrences in speech and 19 in writing). The

influence of medium will be briefly discussed in section 7, but the main analysis will be based on

the total number of constructions (i.e. in speech andandandand writing), following a methodology outlined in

the next section.

10 For the sake of convenience, I will refer to the frequency of the constructions. Note, however, that in some cases a

construction contains more than one non-finite verb, e.g. (i), and that the figures given actually correspond to the

number of non-finite verbs in the constructions.

(i) Well that’ll makemakemakemake them sit up and think that’s for sure.

(KD7 37)

11 This, incidentally, is also the case for causative have, which is usually described in the literature as more formal

(see e.g. Murphy 1985: 92 or Thomson & Martinet 1980: 107).

GAËTANELLE GILQUIN


8

4. The technique of multiple distinctive collexeme analysis4. The technique of multiple distinctive collexeme analysis4. The technique of multiple distinctive collexeme analysis4. The technique of multiple distinctive collexeme analysis

Relying on the notion of collocation and the theoretical framework of Construction Grammar,

Gries and Stefanowitsch have developed a method aimed at investigating the interaction between

words and constructions. The method, known as collostructional analysis, measures the

association strength between a particular construction and the lexemes occurring in a given slot of

this construction.12 The lexemes attracted to a construction are referred to as “collexemes”, and

the combination of a construction and a collexeme is called a “collostruction”. Collostructional

analysis includes three different techniques, viz.

(i) collexeme analysis, which studies one slot in a particular construction, e.g. the V

slot in the [X think nothing of Vgerund] construction;

(see Stefanowitsch & Gries 2003)

(ii) distinctive collexeme analysis, which studies one slot in two or more similar

constructions, e.g. the verb in the ditransitive and to-dative constructions;

(see Gries & Stefanowitsch 2004a)

(iii) covarying collexeme analysis, which studies the interaction between two slots in a

particular construction, e.g. V1 and V2 in the [X V1 Y into V2gerund] causative

construction.

(see Gries & Stefanowitsch 2004b)

The technique that will be used here is that of distinctive collexeme analysis or, more

precisely, multiple distinctive collexeme analysis, as it will involve more than two alternating

constructions.13 In order to qualify for distinctive collexeme analysis, the various constructions

should be functionally and/or semantically near-equivalent. This is clearly the case of the dative

alternation, mentioned above. Periphrastic causative constructions can claim to this status, too.

12 The idea of slot is what distinguishes collostructional analysis from collocational analysis. While collocational

analysis examines all the words occurring in a specified span around the node, collostructional analysis limits its

investigation to the words occurring in a particular slot of a construction. 13 A “simple” distinctive collexeme analysis (i.e. with only two alternatives) will be used in section 7 to determine the

influence of medium on collexemes.



9

They all result from the process of causative formation (Siewierska 1991: 29), by which the

valency of the verb is increased by one. They are often claimed to share the same basic meaning,

namely that of indirect causation (as opposed to lexical causative verbs, which are said to express

more direct causation, see e.g. Baron 1974: 327), and they are sometimes presented as

synonymous (Visser 1973: 2269, for instance, writes that, when used with a causative meaning,

have “might be apprehended as a synonym of make or get to, and consequently is often

replaceable by them”). Finally, although the different structures do not necessarily involve the

same participants at the syntactic level, they all include a CAUSER and a CAUSEE at the conceptual

level.14 Compare:

(3) I [CAUSER] hadhadhadhad Elsie [CAUSEE] go on a Wednesday night.

(KCP 566)

(4) He [CAUSER] hadhadhadhad him [CAUSEE] playing yesterday.

(KBD 8829)

(5) They [CAUSER]’ve just hadhadhadhad a new double glazed back door put on [by CAUSEE].

(KCR 59)

Given this near-equivalence, the technique of multiple distinctive collexemes can therefore

be applied to periphrastic causative constructions.

Let us now illustrate the computation of multiple distinctive collexemes by means of the

example of take (cf. table 4).15 The multiple distinctive collexeme analysis first examines the

observed frequency of the verb in each causative construction (5 occurrences in [X CAUSE Y Vto-

inf], 8 occurrences in [X GET Y Vto-inf], 2 occurrences in [X GET Y Vpp], etc., for a total of 35

14 The PATIENT, by contrast, is optional, only appearing in cases where the non-finite verb is transitive. Compare (4)

with:

(i) You [CAUSER] can’t havehavehavehave Joe [CAUSEE] doing it [PATIENT].

(KST 3266)

15 See also the help files accompanying the program Coll.analysis 3 (Gries 2004).

GAËTANELLE GILQUIN


10

occurrences). On the basis of the total frequency of each construction (see table 3), it then

establishes the expected frequency, that is the frequency that would be expected if the 35

occurrences of take were distributed in proportions matching those of the different constructions.

For each construction, a binomial test is performed in order to establish the probability of a

particular observed frequency given the expected frequency (e.g. the probability to find 5

occurrences of take in [X CAUSE Y Vto-inf] when you would have expected it 2.09 times). This

probability is then log-transformed (=log10 pbinomial value) and gets a positive sign when the verb

occurs more frequently than expected in the construction and a negative sign when the verb occurs

less frequently than expected.16 The resulting value, the distinctiveness value, makes it possible to

determine (i) whether a given verb is distinctive for a particular construction or not (positive

values indicate attracted collexemes, negative values indicate repelled collexemes), and (ii)

whether the co-occurrence between the verb and the construction is statistically significant or not

(the co-occurrence is statistically significant if the absolute distinctiveness value is higher than

1.30103, p<0.05). Take, for instance, appears to be attracted to [X CAUSE Y Vto-inf], [X GET Y

Vto-inf], [X HAVE Y Vinf] and [X HAVE Y Vpp], but repelled by the other constructions (cf.

negative values). The co-occurrence, however, is only statistically significant in the case of [X

GET Y Vto-inf] and [X GET Y Vpp] (absolute distinctiveness value > 1.30103).

16 Note that in their earlier papers, Gries and Stefanowitsch used the p-value as the measure of association strength

between a word and a construction. See Gries et al. (2005) for a study using a log-transformed p-value and why such

a value is preferable to the p-value.



11

taketaketaketake cause

+ to

get

+ to

get

+ pp

get

+ prp

have

+ inf

have

+ pp

have

+ prp

make

+ inf

make

+ to

make

+ pp

observed

frequency

5 8 2 1 2 9 0 8 0 0

expected

frequency

2.09 3.79 7.81 1.26 0.73 6.48 0.69 10.90 0.97 0.29

distinctive-

ness value

1.26 1.51 -2.06 -0.20 0.79 0.73 -0.30 -0.72 -0.43 -0.13

Table 4. Observed frequency, expected frequency and distinctiveness value for take

The multiple distinctive collexeme analysis was performed by means of Coll.analysis 3

(Gries 2004), a program for R for Windows which can be used for the computations necessary for

the different techniques belonging to the family of collostructional analysis (see above). For a

multiple distinctive collexeme analysis, the program requires as input a file listing the

constructions in one column and the verbs occurring in them in the other column, with one line

per occurrence of a construction, as illustrated by (6).17

(6) constructionconstructionconstructionconstruction v_slotv_slotv_slotv_slot

cause_to approach

cause_to be

get_pp clean

get_pp do

get_pp do

get_prp go

get_to buy

Next to the observed and expected frequencies and the distinctiveness value, the output file

also provides a measure called “SumAbsDev”, which is the sum of all absolute distinctiveness

17 Note that it is the lemma of the non-finite verb, not its form, which is used in the analysis. An analysis based on the

form of the verb (e.g. approach, cleaned, going) could only compare those constructions which have the same type of

non-finite complement (e.g. [X GET Y Vpp], [X HAVE Y Vpp], and [X MAKE Y Vpp] for the past participle

complement). Since here we are aiming at a comparison of all the ten causative constructions presented in table 1, all

verb forms have been lemmatised.

GAËTANELLE GILQUIN


12

values for a particular verb. The higher the result, the more strongly the verb deviates from its

expected distribution. Finally, the file specifies the construction which displays the strongest

deviation of observed from expected (LargestDev). This deviation can correspond to the highest

degree of attraction (with a positive distinctiveness value) or the highest degree of repulsion (with

a negative distinctiveness value). In the case of take, SumAbsDev = 8.11 and the strongest

deviation is displayed by [X GET Y Vpp] (repulsion).

The results of a multiple distinctive collexeme analysis will prove particularly insightful in

the case of periphrastic causative constructions, especially given the lack of information available

about the verbs found in the V slot (see section 2) and the fact that these constructions are

regularly regarded as synonyms. On the one hand, the analysis will highlight the verbs that exhibit

a strong preference for one of the ten causative constructions as opposed to the others (i.e. its

collexemes). On the other hand, comparing the constructions’ collexemes and examining them as

for their potential associations with semantic fields will help us identify the distinctive meanings

of the different constructions, thus supporting the argument defended here that each construction

should be considered as a construction in its own right with its own semantics. Before going on to

the presentation and discussion of the results, however, it might be worth considering two

objections that could be raised against an analysis such as the one proposed here, namely the

possibility of using raw frequency counts instead, and the suggestion that a more abstract

construction schema might be preferable.

5. Two5. Two5. Two5. Two possible objections possible objections possible objections possible objections

5.1. Frequency

Here is how Gries et al. (2005: 648) summarise the advantages of collexeme analysis over

frequency-based approaches:



13

(i) collexeme analysis does not neglect the word’s and the construction’s overall

frequencies;

(ii) collexeme analysis allows for identifying cases where a construction and a word

repel each other;

(iii) collexeme analysis allows for separating the wheat from the chaff by

distinguishing significant from random co-occurrence.

The same types of arguments can be invoked in favour of multiple distinctive collexeme

analysis. Besides the frequency of a word in a particular construction, multiple distinctive

collexeme analysis also takes into account the frequency of each of the alternating constructions

and the overall frequency of the word in these constructions. As will be briefly exemplified below,

this can lead to results which are different from, but more accurate than the results produced by an

analysis relying merely on the frequency of the word in the construction. Second, it has already

been noted that, next to the positive distinctiveness values, which show that a word is attracted to

a construction, the computation also produces negative values, which are indicative of repulsion

between the word and the construction. Although in this study the focus will be on the most

distinctive collexemes of the different constructions, and little will be said about the verbs that are

repelled by them, a more detailed analysis could also exploit the relations of repulsion to provide

a more precise description of the constructions. Third, the distinctiveness values can be labelled as

statistically significant or not (cf. threshold value of 1.30103). In the tables of results presented in

section 6, the values are statistically significant18 and the preference of the verbs for a particular

construction can therefore not be attributed to chance.

Let us come back to the first advantage of the multiple distinctive collexeme analysis and

illustrate it with a couple of examples. As will appear from the results for [X MAKE Y Vpp] (table

7), the distinctiveness values and the raw frequencies do not necessarily correspond to each other.

Thus, feel has a raw frequency of 5 and a distinctiveness value of 1.98, whereas understand has a

18 The occasional non-significant values are put between brackets.

GAËTANELLE GILQUIN


14

lower raw frequency of 2 but a higher distinctiveness value of 2.99. This mismatch is due to the

fact that, although feel is more frequent in [X MAKE Y Vpp] than understand, it is also more

frequent in the other constructions (in total, it occurs 156 times, while understand occurs 6 times

only), which makes it less distinctive for [X MAKE Y Vpp]. Bringing into question which

measure, frequency or distinctiveness, is more appropriate, Gries et al. (2005, to appear) argue for

the latter. Using experimental data, they demonstrate that collexeme analysis significantly

outperforms frequency when it comes to predicting speakers’ behaviour, both in terms of

production (sentence-completion tasks) and comprehension (reading tasks). It can reasonably be

assumed that the same is true of multiple distinctive collexeme analysis.

Another case in point is the occurrence of do in the V slot of [X GET Y Vpp] and [X

HAVE Y Vpp] (see table 9). While a frequency-based approach would underline the predominance

of this verb in both constructions, the multiple distinctive collexeme analysis highlights two other

verbs, sort out and cut, as the most distinctive collexemes of [X GET Y Vpp] and [X HAVE Y Vpp]

respectively and thus helps us identify the typical meanings of these constructions more easily.

The verb do still occupies a prominent position, being ranked second and third respectively, but

given that it is frequent in both constructions and in fact occurs in most of the other causative

constructions, too, it does not really distinguish between them and its importance is therefore

slightly less.

5.2. Construction schema

The second possible objection concerns the degree of abstraction of the construction schema.

While this analysis is based on ten constructions, corresponding to each of the four causatives

cause, get, have and make, together with the different non-finite complements they take, one

might wonder whether it would not be more economical to adopt a more abstract schema of the

causative construction, centred around the causative alone. In other words, would we not obtain



15

the same results, but with less computation, if we simply considered the four schemas under (7),

instead of the ten schemas listed in table 1?

(7) [X CAUSE Y V]

[X GET Y V]

[X HAVE Y V]

[X MAKE Y V]

The answer to this question is no. A detailed examination of periphrastic causative

constructions (cf. Gilquin 2004) reveals some common features among the different specific

schemas related to one and the same causative (for instance, the three patterns with get are all

almost exclusively used with an animate CAUSER and tend to involve some sort of difficulty, see

later). However, it also shows that these schemas present major differences which justify making a

distinction between them (the CAUSEE, for example, is usually animate in [X GET Y Vto-inf],

inanimate in [X GET Y Vprp] and unexpressed in [X GET Y Vpp]). The same is true of their

collexemes. As a rule, the different patterns are associated with different collexemes in the V slot.

This is obvious from table 5, which lists, for each pattern with have, the three most distinctive

collexemes and compares their distinctiveness values and ranks with those of the other two

patterns.19 With only two exceptions (in bold), the verbs that are the most distinctive for one

pattern are repelled by the other patterns (cf. negative values). Of course, the repulsion can

sometimes be explained by characteristics of the verb. Thus, come and go cannot be used in the

passive voice and so cannot possibly appear in the past participle construction. Know and believe

are normally not found in the progressive form, which makes their appearance in the present

participle construction unlikely. These characteristics, however, are clearly not the sole

explanation, since certain verbs could theoretically be found in any of the three patterns, but still

19 This table is based on a multiple distinctive collexeme analysis of the have-constructions only, which explains why

the results are different from those presented later, where the analysis was carried out for all the causative

constructions together.

GAËTANELLE GILQUIN


16

show a strong preference for one of them. Thus, the verb do, which can be used both in the active

and passive voice and both as a simple and progressive form, is significantly more distinctive for

the past participle construction, as in (8), than for the infinitive or present participle construction,

cf. (9) and (10).20

(8) Well we’ve got to havhavhavhaveeee ours [= sewing machine] done.

(KCB 384)

(9) They gotta havehavehavehave an electrician do the job.

(KB7 7728)

(10) And of course we hadhadhadhad a machine doing all that work.

(KP1 4928)

[X HAVE Y Vinf] [X HAVE Y Vpp] [X HAVE Y Vprp]

know (1) 7.25 (265) -5.23 (252) -0.28

come (2) 4.84 (266) -5.97 (27) 0.820.820.820.82

believe (3) 3.11 (263) -2.24 (237) -0.12

do (267) -5.07 (1) 4.99 (266) -1.16

cut (266) -2.22 (2) 4.55 (267) -2.10

perm (265) -0.92 (3) 1.89 (265) -0.87

go (46) 0.560.560.560.56 (267) -14.93 (1) 15.06

work (256) -0.29 (264) -5.23 (2) 7.42

go on (241) -0.13 (259) -2.24 (3) 3.18

Table 5. Distinctiveness values and ranks of some verbs in the three have causative patterns

20 Besides this quantitative difference, we also notice a qualitative difference between the three constructions. In (8),

do has a vague meaning and actually replaces the verb fixed. In (10), it is used to refer to the action of an inanimate

CAUSEE (“a machine”). (9), by contrast, has neither of these characteristics. Gilquin (forthcoming) examines the

differences in sense often exhibited by collexemes shared by several constructions.



17

The same differences in attraction/repulsion exist with [X GET Y V] and [X MAKE Y

V].21 From now on, the ten patterns listed in table 1 will therefore be considered as individual

constructions and analysed as such.

6. Results and discussion6. Results and discussion6. Results and discussion6. Results and discussion

6.1. Deviation from expected frequency

For each verb, the multiple distinctive collexeme analysis compares the observed frequency with

the expected frequency, that is the frequency of the verb if it were distributed in proportions

matching those of the different constructions. If, as the literature seems to suggest, there were no

significant differences between the V slots of the various causative constructions, we would not

expect great deviations of observed from expected, since each verb would be distributed in a way

corresponding to the frequencies of the constructions. In the corpus data, however, SumAbsDev,

which measures the extent to which a verb deviates from its expected distribution, has values up

to 122 and indicates potentially significant tendencies up to rank 514 (out of 843 ranks).22 This

shows that the observed frequencies of the verbs across the causative constructions can strongly

diverge from the expected ones and, consequently, that causative constructions do differ in the

filling of their V slots, attracting and repelling essentially different verbs.

Table 6 presents the collexemes with a SumAbsDev value superior to 50, i.e. the verbs

with the strongest overall deviation of observed from expected. It appears that the verb with the

highest SumAbsDev, do, significantly disprefers the [X MAKE Y Vinf] construction. It is, on the

other hand, significantly distinctive for [X GET Y Vpp] and, to a lower degree, [X HAVE Y Vpp].

These two opposite tendencies explain the very high value of SumAbsDev (which, it will be

reminded, takes account of the absoluteabsoluteabsoluteabsolute distinctiveness values). The second verb, go, is highly

21 With cause, only one type of pattern is possible, namely [X CAUSE Y Vto-inf]. 22 After rank 514, SumAbsDev is smaller than 1.30103, the value for p=0.05, so that none of the individual

distinctiveness values can possibly reach the threshold level of significance.

GAËTANELLE GILQUIN


18

distinctive for [X GET Y Vprp], but also occurs significantly in [X HAVE Y Vprp]. Feel, laugh and

look significantly prefer [X MAKE Y Vinf] (with feel also showing a significant distinctiveness

value in [X MAKE Y Vpp]), while sort out is significantly attracted to [X GET Y Vpp]. Verbs with

a smaller SumAbsDev also display such preferences for one or two constructions, which

challenges the view that periphrastic causative constructions are “always safe”, but also

potentially hints at subtle semantic differences.

cause

+ to

get

+ to

get

+ pp

get

+ prp

have

+ inf

have

+ pp

have

+ prp

make

+ inf

make

+ to

make

+ pp

ΣAbsDev LargDev

do -12.83 -1.04 40.17 -5.29 -2.72 13.04 -0.29 -42.63 -2.10 -2.13 122.24 make + inf

go -3.38 0.59 -18.19 62.55 -0.26 -14.75 7.39 -0.71 -0.80 -0.60 109.23 get + prp

feel -3.14 -7.76 -17.09 -1.66 -1.42 -13.86 -1.35 54.71 0.84 1.98 103.81 make + inf

laugh -2.95 -5.47 -12.05 -1.05 -1.00 -9.77 -0.95 53.36 -1.34 -0.40 88.34 make + inf

sort out -2.03 -2.77 44.99 -1.21 -0.69 -6.75 -0.66 -10.76 -0.93 -0.28 71.08 get + pp

look -2.57 -1.35 -10.52 -0.87 -0.87 -8.53 -0.36 32.13 0.30 -0.35 57.85 make + inf

Table 6. Distinctiveness values of the verbs with SumAbsDev > 50

In what follows, we will consider the most distinctive collexemes of each construction. By

examining how they fall into specific semantic fields, it will be possible to provide a description

of the typical meanings of the different causative constructions.23

23 Note that the classification of the collexemes will be based on intuitive assessment of which verbs form coherent

semantic classes. For a more objective procedure, one could cluster the verbs on the basis of their collocates or

collexemes (see Gries & Stefanowitsch, to appear). Alternatively, as suggested by a reviewer, one could use a well-

established classification scheme such as Levin’s (1993) verb classes. Thus, Levin’s class of “Verbs Involving the

Body”, and more precisely “Verbs of Bodily Processes” (Levin 1993: 217ff.), includes most of the verbs which I

describe as expressing physiological processes and which are associated with [X MAKE Y Vinf] (see section 6.3).

Certain classes of verbs which I introduce, however, do not have equivalents in Levin’s classification. This is the case

of the class of verbs referring to the frame of service, found to be associated with [X HAVE Y Vpp] (see section 6.4),

whose members belong to different classes in Levin’s classification scheme (e.g. perm is a Braid Verb, whereas build

belongs to the Build Verbs). Note, moreover, that phrasal verbs are largely absent from Levin’s classification and

would therefore have to be disregarded in the analysis.



19

6.2. [X MAKE Y Vpp]

Let us start with the construction whose V slot receives the most attention in the literature, viz. [X

MAKE Y Vpp]. Table 7 shows the distinctive collexemes for this construction, arranged in

descending order according to their distinctiveness values. Know, illustrated in (11), is the most

distinctive collexeme and the only collexeme with a highly significant value (p<0.001).

(11) TDs who opposed the amendment, such as Oliver J. Flanagan on 13 May and Dr

Michael Woods on 14 May, began to call on the silent majority, to makemakemakemake their

opposition to divorce known.

________________________________________________________________________________________________________________________

Collexeme (n) Distinctiveness Collexeme (n) Distinctiveness

________________________________________________________________________________________________________________________

1 knowknowknowknow (19) 32.22 5 light (1) 1.78

2 understandunderstandunderstandunderstand (2) 2.99 6 recognizerecognizerecognizerecognize (1) 1.48

3 swallow (1) 2.08 7 hearhearhearhear (1) 1.31

4 feelfeelfeelfeel (5) 1.98 (8 pay (1) 0.69)

________________________________________________________________________________________________________________________

Table 7. Distinctive collexemes for [X MAKE Y Vpp]

Among the seven significantly distinctive collexemes (pay does not reach the threshold

level of significance and is only mentioned in table 7 for the sake of completeness), five involve

some sort of influence (in bold in the table), as suggested by Van Ek and Robat (1984: 327). Four

of them, in fact, are explicitly mentioned in the literature, namely known, felt (Van Roey 1982:

84), understood and heard (Berland-Delépine 1990: 173). Contrary to what Van Ek and Robat

(1984) claim, however, the verbs do not denote the exercise, but only the recognition of influence

on the part of the CAUSEE. While exercise of influence is expressed, it is not by the verb itself, but

by the construction as a whole: if the CAUSER makes his or her influence felt, s/he exercises some

influence.

GAËTANELLE GILQUIN


20

Table 7 reveals that 90% of the verbs occurring in [X MAKE Y Vpp] belong to the same

semantic class. This might explain why several linguists have commented on the V slot of [X

MAKE Y Vpp], although the construction itself is relatively infrequent. As will become clear in

what follows, however, the other causative constructions deserve attention, too, for, although their

collexemes do not display as high a degree of cohesion as [X MAKE Y Vpp], their distinctiveness

values are often higher and therefore point to interesting tendencies.

6.3. [X MAKE Y Vinf] and [X BE made Vto-inf]

The [X BE made Vto-inf] construction results from the passivisation of the main clause of [X

MAKE Y Vinf]. This process of passivisation may be thought to have no effect on the V slot itself.

Yet, table 8 shows that, despite some similarities, the two constructions have quite different

collostructional profiles. The most obvious difference lies in the degree of distinctiveness of the

collexemes. While with [X MAKE Y Vinf] the top collexemes have a distinctiveness value of over

50, with [X BE made Vto-inf] it never reaches 5 (of all the causative constructions under

investigation, [X BE made Vto-inf] actually presents the lowest distinctiveness values for its verbs).

In other words, the bare infinitive construction with make displays very strong preferences for a

number of verbs, especially feel and laugh, cf. (12) and (13), whereas the to-infinitive construction

displays a much lower degree of association with its V slot.

(12) Not those kind of relaxants but something just to relieve the tension and makemakemakemake her

feel calmer.

(KBK 469)

(13) I must just tell you this, Laura did makemakemakemake me laugh, cos she said <pause> she stood

up and she said, I’m gonna give my talk about cats <pause> so I said, fine.



21

________________________________________________________________________________________________________________________

[X MAKE Y Vinf] [X be made Vto-inf]

Collexeme (n) Distinctiveness Collexeme (n) Distinctiveness _______________________________________________________________________________________________________________________________________________________

1 feelfeelfeelfeel (142) 54.71 1 mean (3) 4.67

2 laughlaughlaughlaugh (109) 53.36 2 perform (3) 3.69

3 looklooklooklook (86) 32.13 3 pay (5) 3.10

4 thinkthinkthinkthink (66) 24.54 4 work (6) 2.73

5 wonderwonderwonderwonder (31) 15.71 5 dependdependdependdepend (2) 2.64

6 appearappearappearappear (29) 10.71 6 suffersuffersuffersuffer (2) 1.97

7 seemseemseemseem (28) 10.25 7 seemseemseemseem (4) 1.94

8 wantwantwantwant (15) 7.60 8 believebelievebelievebelieve (2) 1.83

9 soundsoundsoundsound (12) 6.08 9 account (1) 1.56

10 jumpjumpjumpjump (13) 4.86 10 adhere (1) 1.56

11 realis|ze realis|ze realis|ze realis|ze (8) 4.05 11 allow (1) 1.56

12 happenhappenhappenhappen (12) 3.86 12 apply (1) 1.56

13 work (26) 3.30 13 argue (1) 1.56

14 meet (8) 3.24 14 ascend (1) 1.56

15 acheacheacheache (6) 3.04 15 begin (1) 1.56

16 ask (6) 3.04 16 bring down (1) 1.56

17 cringecringecringecringe (6) 3.04 17 contribute (1) 1.56

18 coughcoughcoughcough (5) 2.53 18 correspondcorrespondcorrespondcorrespond (1) 1.56

19 last (5) 2.53 19 disassemble (1) 1.56

20 smilesmilesmilesmile (5) 2.53 20 enable (1) 1.56

21 sneezesneezesneezesneeze (5) 2.53 21 encapsulate (1) 1.56

22 worryworryworryworry (4) 2.03 22 failfailfailfail (1) 1.56

23 refer (5) 1.89 23 grasp (1) 1.56

24 crycrycrycry (7) 1.88 24 intensify (1) 1.56

25 wait (7) 1.88 25 oscillate (1) 1.56 _________________________________________________________________________________________________________________________________________________________

Table 8. Most distinctive collexemes for [X MAKE Y Vinf] and [X BE made Vto-inf]

The two constructions also differ in the way the collexemes fall into semantic classes. With

[X MAKE Y Vinf] many of the collexemes share one characteristic, namely they refer to processes

GAËTANELLE GILQUIN


22

where no volition is involved, even in cases where the CAUSEE is animate.24 These collexemes are

in bold in table 8. Different subclasses can be distinguished among these non-volitional verbs, the

most prominent ones being descriptive verbs (look, appear, seem or sound) and verbs describing

mental processes (think, wonder, realise or worry) and physiological processes (ache, cough,

sneeze or cry), as illustrated by (14), (15) and (16) respectively.25

(14) This mademademademade the accident appear reasonable, something which even they could have

done.

(A5Y 1310)

(15) It makesmakesmakesmakes you think of summer dunnit?

(KBG 3234)

(16) Oh poetry <pause> <sigh> some of those poems that Bon had in her Touchstones

book, especially the ones, the war, about the war, I read some of those and it mademademademade

me cry, it was so sad.

(KDM 6595)

By contrast, the collexemes of [X BE made Vto-inf] cannot be said to belong together

semantically in a very obvious way (cf. top collexemes: mean, perform, pay, work). As is the case

for the active construction, the list includes a number of non-volitional verbs (in bold), some of

which are common to [X MAKE Y Vinf] (compare (17) and (18)), but there are fewer of them and

their distinctiveness value is lower (for seem it equals 1.94, as opposed to 10.25 with the active

construction).

24 It is interesting to note that, even verbs like look or jump, which can theoretically be used both volitionally and

non-volitionally (compare He looked through the window and You look tired), turn out to take on a non-volitional

meaning in almost all their occurrences in the construction (see section 7). 25 Of these three subclasses, only the first one, the subclass of descriptive verbs, can also be used with inanimate

CAUSEES, cf. (14).



23

(17) The effect of the device is to remove or reduce human agency, makingmakingmakingmaking events seem

the consequence of impersonal forces such as ideology, the unconscious, history, or

language itself.

(A1A 116)

(18) One law for the rich and another for the poor, as the two systems can be mademademademade to

seem, are laid down together in a book which commemorates a desertion, on the

author’s part, of the rich for the poor.

(A05 265)

If we compare these results with what the literature says about the V slot of causative

constructions with make, we notice that in the case of [X MAKE Y Vinf], Van Ek’s (1966)

description (see section 2) is quite accurate, since the verbs he mentions (feel, look, seem, think,

laugh and smile) all belong to the top twenty collexemes. It should be emphasised, however, that

Van Ek’s study, like this one, is corpus-based and hence represents an exception in the literature

on causative constructions rather than the rule. As for the passive construction, its absence from

the literature may be due to the lower distinctiveness values of its collexemes and the lack of clear

semantic classes among them or, more simply, to the assumption that the passive construction has

the same lexical preferences as its active counterpart. More generally, it appears that the common

description of make as a causative expressing coercion (see e.g. Werner et al. 1990: 392) is

inaccurate. Even if we restrict ourselves to constructions with animate CAUSEES, the presence of a

non-volitional verb rules out a coercive interpretation, for one cannot possibly force a person to do

something which is not dependent on their will. Instead, make seems to primarily express the

causation of a process that is not directly dependent on the CAUSEE, a meaning which one would

certainly not expect from a verb which is often regarded as the most general and prototypical

causative (cf. Dixon 2000; Altenberg 2002).

Get and have, too, appear to be misrepresented in the literature, as we will see presently.

While they are often described as (near) synonyms, it will be demonstrated that the get and have

constructions have essentially different collostructional profiles.

GAËTANELLE GILQUIN


24

6.4. [X GET Y Vpp] and [X HAVE Y Vpp]

A comparison between the first and second half of table 9 reveals that the collexemes of [X GET

Y Vpp] and [X HAVE Y Vpp] are quite different. Not only do the latter exhibit weaker

distinctiveness, but the verbs themselves are different. Apart from the verb do, which is distinctive

for both constructions (compare (19) and (20)) – although more so for the get construction – there

is no overlap between the most distinctive collexemes of the two constructions, which is a clear

indication that the constructions are not interchangeable.

(19) They [= the solicitors]’re trying their best to getgetgetget it [= the contract] done before

Christmas.

(KB7 14832)

(20) Tomorrow would be quite nice because I’m havinghavinghavinghaving my hair done tomorrow.

(KBK 2321)



25

__________________________________________________________________________________________________________________________

[X GET Y Vpp] [X HAVE Y Vpp]


__________________________________________________________________________________________________________________________

1 sort outsort outsort outsort out (74) 44.99 1 cutcutcutcut (53) 24.95

2 do (278) 40.17 2 permpermpermperm (22) 16.12

3 sortsortsortsort (16) 10.43 3 do (183) 13.04

4 finishfinishfinishfinish (17) 8.33 4 buildbuildbuildbuild (16) 8.99

5 organis|zeorganis|zeorganis|zeorganis|ze (15) 7.85 5 cutcutcutcut offoffoffoff (10) 7.33

6 dress (9) 5.87 6 put downput downput downput down (10) 7.33

7 wash (13) 5.50 7 put (17) 7.09

8 fixfixfixfix (14) 5.14 8 serviceserviceserviceservice (8) 4.99

9 book (7) 4.56 9 cleancleancleanclean (11) 4.45

10 start (11) 3.08 10 remove (7) 3.73

11 dry (4) 2.61 11 put upput upput upput up (6) 3.63

12 write down (5) 2.57 12 shaveshaveshaveshave (6) 3.63

13 deliver (7) 2.42 13 test (8) 3.48

14 do up (4) 1.99 14 put input input input in (9) 3.37

15 insure (4) 1.99 15 put on (8) 3.14

16 set upset upset upset up (4) 1.99 16 replace (5) 2.96

17 cook (3) 1.96 17 trimtrimtrimtrim (5) 2.96

18 line up (3) 1.96 18 repairrepairrepairrepair (7) 2.92

19 send off (3) 1.96 19 take out (5) 2.49

20 warm up (3) 1.96 20 make (14) 2.22

21 work outwork outwork outwork out (3) 1.96 21 check out (3) 2.20

22 kill (4) 1.60 22 hhhhighlightighlightighlightighlight (3) 2.20

23 make up (3) 1.43 23 knock downknock downknock downknock down (3) 2.20

24 backdate (2) 1.30 24 put across (3) 2.20

25 blow (2) 1.30 25 shapeshapeshapeshape (3) 2.20

__________________________________________________________________________________________________________________________

Table 9. Most distinctive collexemes for [X GET Y Vpp] and [X HAVE Y Vpp]

Let us now try to determine the semantic values of the constructions on the basis of their

collexemes, starting with the have construction. Most of the collexemes listed for [X HAVE Y

GAËTANELLE GILQUIN


26

Vpp] share one important feature, namely they are related to the frame of service: a hairdresser

cutting, perming, trimming, highlighting or shaping a customer’s hair, a builder building a house

or knocking down a wall for someone, a vet putting down an animal, a mechanic servicing

somebody’s car, etc. Note that the verb do is very often used in [X HAVE Y Vpp] with reference

to hairdressing, as in (20) above. The typical meaning of [X HAVE Y Vpp], illustrated in (21) and

(22), can therefore be described as ‘to commission someone to do something’.

(21) Well she’s growing the back of it <pause> and havinghavinghavinghaving it permed and highlighted.

(KCE 4006)

(22) I mean if if if you’ve hadhadhadhad your vehicle serviced and the sump plug hasn’t been put

back in, then obviously you’ve got a perfectly legitimate claim against the person

that’s done the work.

(KRL 773)

The list of distinctive collexemes for [X GET Y Vpp] is more difficult to classify

semantically. Two semantic fields seem to emerge from table 9, viz. one having to do with

organisation (e.g. sort (out), finish, organis|ze) and the other describing daily actions (e.g. dress,

wash, cook). Compare:

(23) And did you getgetgetget everything sort of organized?

(KBC 5132)

(24) I thought right get up and getgetgetget them bloody curtains washed.

(KCG 2648)

In order to understand how these collexemes relate to the general meaning of [X GET Y

Vpp], it may be useful to examine other features of the construction, more particularly its semantic

prosody26 and the nature of its CAUSEE (see Gilquin 2004). An investigation of the words

occurring in the immediate environment of causative get reveals that many of them imply some

26 Semantic prosody refers to the tendency for words to collocate with other words from a definable semantic set (see

e.g. Stubbs 1995a, b).



27

sort of effort or difficulty, cf. difficult and deserves credit in (25).27 Besides, it is not uncommon to

find a reference to a deadline, as in (26), emphasising some urgency.

(25) It is easy to criticise the government’s decisions: after a century of inertia, making

changes in London was always going to be difficultdifficultdifficultdifficult, and Mrs Bottomley deserves deserves deserves deserves

creditcreditcreditcredit for gettinggettinggettinggetting the process started.

(FT1 159)

(26) He told me it was still there and that he’d been told he’s got to write to them today

<pause> and they’ve got to getgetgetget it [= the caravan] taken off within twentywithin twentywithin twentywithin twenty eight eight eight eight

daysdaysdaysdays.

(KCN 4680)

This explains the presence of verbs such as sort out (which has a distinctiveness value as

high as 44.99) or finish. Sort out implies the existence of a problem and finish, by referring to the

final stage of an action, is compatible with an urgent deadline. Compare (27) and (28).

(27) We’ll trytrytrytry and getgetgetget something sorted out before you babysit on the <pause>

nineteenth.

(KC8 81)

(28) I bet I’m not gonna getgetgetget this finished in timein timein timein time now.

(KDB 224)

As for the presence of verbs describing daily actions, they can be explained by the frequent

identity between the CAUSER and the (mostly implied) CAUSEE of [X GET Y Vpp], a phenomenon

which represents 63% of all the occurrences of the construction,28 e.g.

27 This, incidentally, is true of all three causative constructions with get. 28 This phenomenon, by contrast, is rare with the other causative constructions (11.70% with [X HAVE Y Vpp], 3.03%

with [X MAKE Y Vpp], 2.91% with [X MAKE Y Vinf] and 0.76% with [X GET Y Vprp]). An example with [X MAKE

Y Vinf] would be:

(i) I didn’t really want to spare the time because you could imagine that I wanted to get ready to come

away <pause> but I mademademademade myself sit and really give him time.

(KBF 952)

GAËTANELLE GILQUIN


28

(29) I’m gonna get, get in touch with her on Saturday <pause> then I’ll getgetgetget my

geography project done, I can’t do anything until I’ve got this bloody project out the

way, can I?

(KCE 6364)

In other words, getting something done often implies that one does it oneself. It is

therefore only natural that verbs describing daily actions should be distinctive for [X GET Y Vpp]

since such actions are typically carried out by the CAUSER him- or herself, e.g.

(30) He won’t eat dumplings, if he sees dumplings in a stew he’ll puke. (…) So I’ve gotta

getgetgetget them cooked before Terry comes home.

(KCX 4535, 4538)

Putting all these elements together, we can summarise the main meaning of the

construction as ‘to carry out an action in difficult circumstances or under a tight schedule’.

It will be reminded that Francis et al. (1996: 306) provide a list of typical examples of the

two constructions just reviewed. It now turns out, however, that these examples mainly describe

services and are therefore more typical of have than of get. This does not mean, of course, that

such verbs are not possible with get. In fact, the collexeme analysis shows that they are possible

but not probable (they do not belong to the top collexemes), hence not distinctive for the get

construction. They may be found in cases where the presence of get is required for other reasons,

especially reasons of semantic prosody. Compare (31) and (32). Although both of them contain

the same verb (cut) and refer to the frame of service, (32) suggests difficulty in carrying out the

action, as shown by the elements in bold italics, and therefore the get construction is preferred.

(31) I just told them you’d hadhadhadhad your hair cut really short.

(KC2 3072)



29

(32) - Would you mind having the boyshaving the boyshaving the boyshaving the boys for about half an hour because I would like to get get get get

Bryony’s hair cut. I want about three or four inches off the bottom (…) and I think

she’s gonna have an absolute screaming fitan absolute screaming fitan absolute screaming fitan absolute screaming fit.

- <laugh> Where’re you taking her?

- Probably to the one at Staffhill if I could have an, if I can get an appointment if I can get an appointment if I can get an appointment if I can get an appointment

boobooboobookedkedkedked.

(KB8 8170 ff.)

In summary, the examination of the collexemes of [X GET Y Vpp] and [X HAVE Y Vpp]

has brought to light a sharp contrast between the two constructions. While the construction with

get involves some sort of difficulty, the construction with have refers to a service, that is an action

which tends to be taken for granted (cf. Duffley 1992: 72) and is normally unproblematic. This

contrast can also be expressed in terms of Hollmann’s (2003: 78ff.) “sphere of control”, which he

presents as characteristic of have. With have, the CAUSER has the CAUSEE in his/her sphere of

control and can therefore apply his/her authority in order to bring about the caused event. With

get, on the other hand, no such sphere of control exists and, with no authority to appeal to,

bringing about the caused event may prove more difficult. This characterisation actually applies to

all the constructions with get and have, but with the other patterns (present participle and infinitive

constructions), the difference is not immediately obvious from the list of collexemes and therefore

little will be said about it in the next two sections.

6.5. [X GET Y Vprp] and [X HAVE Y Vprp]

In terms of collexemes, the present participle construction is certainly the construction that is the

most alike between get and have. Table 10 shows that the top collexeme in both [X GET Y Vprp]

and [X HAVE Y Vprp] is go, as in (33) and (34). The distinctiveness value of this verb, however,

is considerably higher with get than with have (62.55 vs. 7.39), which can probably be related to

the special status of the get construction with going, which is often regarded as an idiom (see e.g.

Kirchner 1952: 225).

GAËTANELLE GILQUIN


30

(33) Yeah, probably if you want me to getgetgetget that old mower going I ought to go up to

Woods and <pause> see if I can get a new drive belt.

(KCH 523)

(34) Yes because y– if you’ve used all your hot water <pause> you can’t havehavehavehave that

boiler going <pause> for an hour or two can you?

(KBF 3246)

More importantly, the verb go points to a class of verbs distinctive for both constructions,

namely verbs of motion (in bold in table 10), which confirms the claim made by Van Ek (1966:

78) about [X GET Y Vprp] (see section 2). Although this finding gives some insight into the use of

the present participle construction, it can only be fully understood if combined with a piece of

information drawn from an investigation of the nature of the CAUSEE in such constructions, which

reveals that the CAUSEE is often inanimate.29 In other words, the most distinctive collexemes of the

construction refer to the motion of objects – either literal motion (cf. a mower, a ball or a car) or

metaphorical motion (ACTION IS MOTION) with e.g. boilers, dishwashers or televisions.

29 Inanimate CAUSEES represent a proportion of 63% in [X GET Y Vprp] and 44% in [X HAVE Y Vprp]. They are, on

the other hand, less frequent in (to-) infinitive constructions, with a percentage of 8% with get and 22% with have

(see Gilquin 2004).



31

________________________________________________________________________________________________________________________

[X GET Y Vprp] [X HAVE Y Vprp]


________________________________________________________________________________________________________________________

1 gogogogo (76) 62.55 1 gogogogo (17) 7.39

2 runrunrunrun (5) 4.03 2 go ongo ongo ongo on (3) 4.52

3 talk (5) 3.33 3 work (7) 4.45

4 movemovemovemove (4) 2.90 4 hang (2) 3.41

5 vote (2) 2.89 5 stick out (2) 3.41

6 stand (3) 2.83 6 play (2) 2.12

7 come income income income in (3) 2.21 7 cry (2) 1.81

8 call (2) 1.75 8 admire (1) 1.71

9 comecomecomecome (4) 1.56 9 dig out (1) 1.71

10 ring in (1) 1.44 10 drill (1) 1.71

11 walk outwalk outwalk outwalk out (1) 1.44 11 go acrossgo acrossgo acrossgo across (1) 1.71

(12 face (1) 1.15) 12 guard (1) 1.71

(13 work (4) 1.05) 13 iron (1) 1.71

(14 lie (1) 0.98) 14 panic (1) 1.71

(15 sing (1) 0.98) 15 roll aroundroll aroundroll aroundroll around (1) 1.71

_____________________________________________________________________________________________________________________

Table 10. Most distinctive collexemes for [X GET Y Vprp] and [X HAVE Y Vprp]

Despite this common meaning of ‘setting an object in motion’, [X GET Y Vprp] and [X

HAVE Y Vprp] display two major differences. First they differ quantitatively, with the get

construction exhibiting higher distinctiveness values (cf. go) and more collexemes belonging to

the semantic class of motion. Second they differ in their semantic prosody, as already shown for

the past participle construction. This appears from a comparison of (33) and (34) above, where the

get construction has an undertone of difficulty (cf. old mower, see if I can) which the have

construction does not have.

It should be added, finally, that the two constructions share another semantic class of

collexemes, viz. verbs of position (in italics in table 10), e.g.

GAËTANELLE GILQUIN


32

(35) I gotgotgotgot the tiles <pause> lying by the side of the hob, but they haven’t yet been cut to

fit <pause> erm <pause> and glued down.

(KBF 1306)

(36) Well you’d have to havehavehavehave it [= the fridge] sticking out.

(KB7 1660)

These verbs, however, are less numerous and their distinctiveness is weaker than that of

verbs of motion (the maximum distinctiveness value, that of [X HAVE Y hanging/sticking out], is

equal to 3.41). In addition, such collostructions tend to belong to the periphery of the category of

causative constructions and actually come close to so-called “existential constructions” (see Quirk

et al. 1985: 1411). Compare (36) with the following, existential sentence taken from Quirk et al.

(ibid.):

(37) I havehavehavehave two buttons missing on my jacket.

6.6. [X GET Y Vto-inf] and [X HAVE Y Vinf]

Generally speaking, the infinitive constructions with get and have exhibit less clear associations

with the V slot than the past participle and present participle constructions. The collexemes have

lower distinctiveness values and cannot be said to cluster around one or two main semantic

classes, as appears from table 11. However, what is striking in the list of collexemes is that it

includes a number of non-action verbs (in bold in the table). For [X GET Y Vto-inf] we notice two

verbs of communication (talk and say) and two verbs of agreement (agree and comply), cf.

(38) But if you can getgetgetget him to say you know, Sting’s deserted me <pause> you know

he, he lets me live on a, you know, my twenty three quid a week pension.

(KC6 1189)

(39) Even so, the difficulty for the individual creating a root definition is less than the

difficulty in gettinggettinggettinggetting all the individuals involved to agree on the definition to be used.

(HRK 428)



33

These collexemes are not the most distinctive ones,30 but because they somehow belong

together, we can tentatively say that the construction is associated with the ‘elicitation of words or

agreement’.

________________________________________________________________________________________________________________________

[X GET Y Vto-inf] [X HAVE Y Vinf]


________________________________________________________________________________________________________________________

1 get (14) 10.82 1 knowknowknowknow (7) 5.76

2 help (10) 9.65 2 come (6) 4.34

3 talktalktalktalk (10) 5.09 3 believebelievebelievebelieve (3) 3.53

4 come (12) 4.14 4 go out (2) 1.85

5 sign (5) 4.09 5 accompany (1) 1.68

6 saysaysaysay (11) 3.89 6 challenge (1) 1.68

7 agreeagreeagreeagree (3) 2.90 7 convert (1) 1.68

8 pick up (4) 2.76 8 disbelievedisbelievedisbelievedisbelieve (1) 1.68

9 give (6) 2.47 9 follow (1) 1.68

10 come in (5) 2.41 10 imagineimagineimagineimagine (1) 1.68

11 provide (3) 2.33 11 insert (1) 1.68

12 stay (4) 2.17 12 inspect (1) 1.68

13 carry (3) 1.97 13 interview (1) 1.68

14 come along (2) 1.93 14 practice (1) 1.68

15 complycomplycomplycomply (2) 1.93 15 put through (1) 1.68

_________________________________________________________________________________________________________________________

Table 11. Most distinctive collexemes for [X GET Y Vto-inf] and [X HAVE Y Vinf]

Likewise, [X HAVE Y Vinf] has the following non-action collexemes: know, believe,

disbelieve and imagine, e.g.

30 The most distinctive collostruction combines the causative and lexical uses of get, cf.

(i) She said I’ll getgetgetget Sally to get me something before Christmas.

(KBE 7832)

GAËTANELLE GILQUIN


34

(40) The bulletin is not a wide ranging, objective, scientific review as De Melker would

havehavehavehave us believe.

(FSY 1369)

(41) I’m nine two this morning on the scales, I’ll havehavehavehave you know.

(KD0 408)

In fact, these verbs belong to the same class of non-volitional verbs that were shown to be

typical of [X MAKE Y Vinf] and [X BE made Vto-inf]. The difference lies in the fact that the

constructions with have are limited to mental verbs and tend to be more idiomatic. The three

instances of believe are used with the auxiliary would (as is the verb imagine) and the pronoun us,

as in (40). As for know, it always occurs in the fixed expression I’ll have you know, cf. (41).

Make, on the other hand, allows a much wider variety of expressions (including CAUSEE-less

constructions with believe). Compare (40) with the following sentences:

(42) It had been such a good idea and all that had resulted from it was a double

punishment for her and a complete failure to makemakemakemake anyone believe Alicia or Daryl

had played the trick.

(KCD 2647)

(43) This “inverted causation” as it has been called, which is a major element of Marxist

theory, is to be found in the theory of Asiatic production, in that the subjects of the

Asiatic despot are mademademademade to believe that they can live because of the blessing of the

god-king, the true guardian and shepherd of the community, while really it is he who

is living off them.

(A6S 535)

(44) This is direct experience, but it is not drama – not until there is some pretence

involved, some symbolic representation, some intention to makemakemakemake believe.

(AM6 126)

[X HAVE Y Vinf], therefore, seems to be associated with the ‘elicitation of a mental

response’, but mainly in idiomatic uses.



35

6.7. [X CAUSE Y Vto-inf]

Let us end our review of the V slot of causative constructions with a construction whose non-

finite complement is never described in the literature in terms of lexical preferences, namely [X

CAUSE Y Vto-inf]. Table 12 shows that the collexemes of this construction exhibit relatively low

distinctiveness values. Yet, they display some semantic cohesion in that two of its most distinctive

collexemes are copular verbs, viz. be (distinctiveness value = 5.61) and become (distinctiveness

value = 4.84), both of which evoke a transformation of the CAUSEE, cf.

(45) The M sixty two in west Yorkshire near junction twenty nine, the loft house

interchange, there are various roadworks and lane restrictions on the slip roads there,

and that’s causingcausingcausingcausing traffic to be slow moving around that junction at the moment.

(HUV 407)

(46) This effectively reduces the file packing, and may also causecausecausecause fixed lengths to become

variable in length.

(FPG 515)

________________________________________________________________________________________________________________________


________________________________________________________________________________________________________________________

1 bebebebe (8) 5.61 11 deposit (2) 2.45

2 rise (4) 4.89 12 differ (2) 2.45

3 becomebecomebecomebecome (5) 4.84 13 evolve (2) 2.45

4 collapse (3) 3.67 14 overestimate (2) 2.45

5 reach (3) 3.67 15 slow down (2) 2.45

6 fall (4) 2.90 16 suffer (3) 2.43

7 see (5) 2.72 17 describe (2) 1.99

8 apprehend (2) 2.45 18 generate (2) 1.99

9 approach (2) 2.45 19 separate (2) 1.99

10 break up (2) 2.45 20 increase (2) 1.99

____________________________________________________________________________________________________________________

Table 12. Most distinctive collexemes for [X CAUSE Y Vto-inf]

GAËTANELLE GILQUIN


36

These two verbs, in fact, belong to a small group of non-volitional verbs, together with see

and suffer, e.g. (47), which suggests that the classic example cause to die may not be such an

improbable combination after all.31

(47) Not long afterwards, his attention fixed on the sufferings of those Poles who had

causedcausedcausedcaused their Jews to suffer, Mendel falls silent, thinking: “Not only us.”

(A05 1531)

Examining the complete list of collexemes of [X CAUSE Y Vto-inf] (i.e. beyond the twenty

collexemes shown in table 12) also reveals the presence of a number of verbs of motion, as was

shown to be the case for the present participle construction with get and have. With cause,

however, the verbs are more specific (e.g. rise, collapse, fall, gyrate, swing round, collide or flow,

to be compared with going or coming for [X GET Y Vprp] and [X HAVE Y Vprp]) and they often

refer to a scientific context, which confirms the stylistic preference of this verb for scientific

genres (cf. Gilquin 2004; see also Chuquet & Paillard 1987: 170). A typical example of the

construction would be:

(48) This cam is in turn pinned to the upper arm of the bell crank C. The resulting

rotation of the crank about its fixed pivot causescausescausescauses the horizontal portion to rise and

fall, lifting link member D and with it the vertically constrained rod member E. The

needle carried in a collet at the lower end is thus made to oscillate vertically.

(FE6 1625)

31 The collexeme die has a (non-significant) distinctiveness value of 1.21 in [X CAUSE Y Vto-inf]. It occurs twice in

the data, once with animate participants and once with inanimate participants:

(i) Using unsafe desensitisation, conventional allergists and general practitioners have causedcausedcausedcaused 26

patients to die since 1957.

(EC7 1596)

(ii) For example, the neocortex is connected to the thalamus in such a way that destruction of part

of the cortex causescausescausescauses cells in a corresponding part of the thalamus to die, a process known as

“Retrograde degeneration”.

(CMH 797)



37

Grouping together the copular verbs be and become and the verbs of motion, we can

conclude that [X CAUSE Y Vto-inf] is most distinctively used to express the process through which

a transformation or a specific movement is caused.32

7. Additional parameters7. Additional parameters7. Additional parameters7. Additional parameters

It was demonstrated in section 5.2 that the form of the V slot is crucial in determining the

collexemes of causative constructions, and hence their meanings. The whole analysis was

therefore carried out on the basis of the more specific schemas, one for each type of non-finite

complement occurring with the different causatives. Other factors, however, appear to influence

the collostructional profile of a construction, too. In the case of causative constructions, these

factors include (but may not be limited to) medium and/or text type, inflection of the causative

and sense of the non-finite complement. For lack of space, these factors will only be briefly

discussed, but each of them would be worth exploring in more depth.

As a rule, the association strength between verbs and constructions seems to be stronger in

speech than in writing. Thus, the collexeme most strongly attracted to [X HAVE Y Vpp] in speech

has a value almost three times higher than its equivalent in writing.33 The difference between the

two media is not only quantitative, but also qualitative, with the spoken and the written [X HAVE

Y Vpp] construction presenting different collexemes. While the top two collexemes in speech are

do and cut, in writing they are baptis|ze and recognis|ze.34 None of these collexemes ever occur in

32 This description may remind one of Langacker’s (1991: 13) “billiard-ball model”, which views the world as

“populated by discrete physical objects (…) capable of moving about through space and making contact with one

another” (emphasis added). It should be noted, however, that this type of causation is hardly ever expressed in

English by means of a periphrastic causative construction, neither with cause nor with any other of the causatives

investigated here (see Gilquin 2006). 33 These results have been obtained by means of a “simple” distinctive collexeme analysis comparing [X HAVE Y

Vpp] in speech and writing. For more details about the computation of this analysis, see Gries & Stefanowitsch

(2004a). 34 Note that a verb like baptis|ze can still be regarded as belonging to the frame of service, which was shown to be

typical of the construction.

GAËTANELLE GILQUIN


38

the other medium. In fact, only 13 out of the 201 collexemes of the construction are shared by

both media (i.e. 6.5%).

In addition, it turns out that the different inflected forms of the causative may have

distinctive collexemes in the V slot. Table 13 illustrates this phenomenon for [X MAKE Y Vinf].

The results are particularly striking for the verb wonder, which is highly significantly preceded by

the form makes, rather than made, make or making (distinctiveness value of makes = 14.29,

p<0.001; SumAbsDev = 24.02). In other words, a sentence such as (49) is far more probable

than one such as (50).

(49) You know it makes you <pause> well it makesmakesmakesmakes you wonder whether <pause>

something could be done without placing him, for his sake!

(KBG 40)

(50) The new titles limit the poems to the field of moral tracts, and makemakemakemake one wonder

exactly what Wordsworth thought his poems were about.

(CAW 1079)

In the same way, we can see from table 13 that work and do are significantly preceded by

make, laugh and think by makes, cry, have and jump by made, and pay by making. All these

differences seem to confirm the idea that “each distinct form is potentially a unique lexical unit”

(Sinclair 1991: 8).35 They also bring to light what could be prefabricated chunks, cf. (it) makes

you wonder above.

35 See Newman & Rice (2006) for an application of this idea to the collocations of eat and drink.



39

________________________________________________________________________________________________________________________

Collexeme made make makes making SumAbsDev LargestDev

_____________________________________________________________________________________________________________________________

1 wonder -2.87 -5.65 14.29 -1.22 24.02 makes

2 work -1.55 4.32 -1.76 -0.48 8.12 make

3 do -0.95 2.45 -3.12 1.04 7.56 makes

4 laugh 0.27 -1.42 3.22 -1.90 6.82 makes

5 think 0.29 -2.76 3.13 -0.31 6.50 makes

6 seem 0.81 -3.31 1.01 1.05 6.18 make

7 cry 3.53 -1.01 -0.94 -0.28 5.76 made

8 have 3.01 -1.23 -1.07 0.29 5.60 made

9 pay -1.11 0.31 -0.88 2.64 4.94 making

10 jump 2.26 -0.39 -0.99 -0.51 4.14 made

________________________________________________________________________________________________________________________

Table 13. Multiple distinctive collexeme analysis for [X MAKE Y Vinf] according to the form of

MAKE

Finally, there is a good case for arguing that verb sense should ideally be taken into

account in a collostructional analysis (see also Gries & Stefanowitsch 2004a: 125). One example

should suffice to illustrate this. The verb look occurs both in [X GET Y Vto-inf] and [X MAKE Y

Vinf], but tends to be used differently. With make it is normally used non-volitionally as a

descriptive verb, cf.

(51) Your hair’s different and it makesmakesmakesmakes you look completely different.

(KC7 25)

With get, by contrast, this sense never occurs in the data. Instead, look is used as a

volitional and abstract process, synonymous with consider, e.g.

(52) The therapist tried to getgetgetget her to look at such situations from her parents’ viewpoint.

(B30 749)

GAËTANELLE GILQUIN


40

These two senses, it will be noticed, confirm the tendencies outlined above (non-volitional

causation with make and potentially difficult process with get, cf. tried to in (52)), which suggests

that taking verb sense into account will not invalidate the results of the analysis, but on the

contrary, will make it possible to improve them, as confirmed by Gilquin (forthcoming). More

generally, it appears that parameters such as those introduced in this section help refine the

description of constructions and should therefore be considered when carrying out a

collostructional analysis.

8. Conclusion8. Conclusion8. Conclusion8. Conclusion

The multiple distinctive collexeme analysis carried out on English periphrastic causative

constructions has demonstrated that, far from being “always safe”, these constructions actually

show very strong preferences for particular (groups of) verbs in the non-finite V slot. In addition,

these collexemes tend to be different for the ten constructions investigated, which argues for

treating each causative construction as a construction in its own right, rather than describing them

as (near) synonyms, as is often the case in the literature.

By examining how the main collexemes fall into semantic classes, it has also been possible

to establish, with various degrees of confidence, a meaning distinctively associated with each

construction, as summarized in table 14. It goes without saying that a closer examination of allallallall the

collexemes (including collexemes with a lower distinctiveness value,36 collexemes not belonging

to a particular semantic class of verbs, as well as collexemes repelled by the construction) would

reveal other, less typical meanings, and would therefore help provide a better description of all the

uses of causative constructions.

36 This was done for the cause construction. The issue has already been raised with respect to collocations. Stubbs

(1995a: 249) notes that “[w]hat is significant is the summed frequency of semantically related items”, and to this,

even less frequently occurring words can contribute.



41

ConstructionConstructionConstructionConstruction Distinctive MeaningDistinctive MeaningDistinctive MeaningDistinctive Meaning

[X CAUSE Y Vto-inf] ‘To cause a transformation or specific movement’

[X GET Y Vto-inf] ‘To elicit words or agreement’

[X GET Y Vpp] ‘To carry out an action in difficult circumstances or under a tight

schedule’

[X GET Y Vprp] ‘To set an object in motion, usually with difficulty’ [also

metaphorical]

[X HAVE Y Vinf] ‘To elicit a mental response’ [mainly idiomatic]

[X HAVE Y Vpp] ‘To commission someone to do something’

[X HAVE Y Vprp] ‘To set an object in motion’ [also metaphorical]

[X MAKE Y Vinf] ‘To cause a process that is not directly dependent on the CAUSEE’

(especially impressions, mental and physiological processes)

[X be made Vto-inf] ‘To cause a process that is not directly dependent on the CAUSEE’

[less marked]

[X MAKE Y Vpp] ‘To exercise some sort of influence’

Table 14. Meanings distinctively associated with periphrastic causative constructions

Moreover, it has been claimed that multiple distinctive collexeme analysis can be refined

by taking other factors into account such as text type, inflection or verb sense. All these

parameters have been shown (if only briefly) to have some quantitative and qualitative influence

on the results and would therefore deserve more attention than they have been given so far in

collostructional analysis.

Finally, this study is an illustration of how multiple distinctive collexeme analysis can be

complemented by other methods, be it from the family of collostructional analysis (cf. “simple”

distinctive collexeme analysis comparing [X HAVE Y Vpp] in speech and writing) or from other

types of analyses (e.g. semantic prosody of the construction or nature of its participants). Such

additional methods, by examining the data from a different (but complementary) angle, help

interpret the results of the multiple distinctive collexeme analysis, thus making the technique even

more powerful.

GAËTANELLE GILQUIN


42

AcknowledgementsAcknowledgementsAcknowledgementsAcknowledgements

Thanks are due to Stefan Th. Gries for providing me with Coll.analysis 3 and helping me

understand the process behind it, as well as to Doris Schönefeld and two anonymous reviewers for

their feedback, which made it possible to substantially improve the paper. Any remaining

shortcomings are solely mine.



43

ReferencesReferencesReferencesReferences

Altenberg, Bengt (2002). Causative constructions in English and Swedish: A corpus-based

contrastive study. In Altenberg, Bengt & Sylviane Granger (eds.) Lexis in Contrast. Corpus-

based Approaches. Amsterdam/Philadelphia: John Benjamins. 97-116.

Baron, Naomi S. (1974). The structure of English causatives. Lingua 33: 299-342.

Berland-Delépine, Serge (1990). Grammaire pratique de l’anglais. Nouvelle édition. Paris: Ophrys.

BNC (2000). British National Corpus. World Edition. Further information available at

http://www.natcorp.ox.ac.uk accessed 27 February 2006.

BNCweb 2.0 (2002). BNCweb: A web-based interface to the British National Corpus. Version 2.0.

Zurich: Zurich University. Further information available at

http://homepage.mac.com/bncweb/home.html accessed 27 February 2006.

Chuquet, Hélène & Michel Paillard (1987). Approche linguistique des problèmes de traduction

anglais – français. Edition révisée. Gap: Ophrys.

Deutschbein, Max & Hermann Klitscher (1959). Grammatik der Englischen Sprache, Auf

wissenschaftlicher Grundlage. Heidelberg: Quelle & Meyer.

Dixon, Robert M.W. (2000). A typology of causatives: form, syntax and meaning. In Dixon,

Robert M.W. & Alexandra Y. Aikhenvald (eds.) Changing Valency. Case Studies in

Transitivity. Cambridge: Cambridge University Press. 30-83.

Duffley, Patrick J. (1992). The English Infinitive. London/New York: Longman.

Eckersley, Charles Ewart & John Macaulay Eckersley (1967). A Comprehensive English

Grammar. London: Longman.

Francis, Gil, Susan Hunston & Elizabeth Manning (1996). Collins COBUILD Grammar Patterns

1: Verbs. London: HarperCollins Publishers.

Gilquin, Gaëtanelle (2004). Corpus-based Cognitive Study of the Main English Causative Verbs.

A Syntactic, Semantic, Lexical and Stylistic Approach. Unpublished PhD Thesis. Louvain-la-

Neuve: Université catholique de Louvain.

GAËTANELLE GILQUIN


44

Gilquin, Gaëtanelle (2006). The place of prototypicality in corpus linguistics. Causation in the hot

seat. In Gries, Stefan Th. & Anatol Stefanowitsch (eds.) Corpora in Cognitive Linguistics:

Corpus-Based Approaches to Syntax and Lexis. Berlin/New York: Mouton de Gruyter. 159-

191.

Gilquin, Gaëtanelle (forthcoming). Making sense of collostructional analysis. On the interplay

between verb senses and collexemes. To appear in Constructions.

Gries, Stefan Th. (2004). Coll.analysis 3. A program for R for Windows 2.x. Available at

http://people.freenet.de/Stefan_Th_Gries accessed 27 February 2006.

Gries, Stefan Th., Beate Hampe & Doris Schönefeld (2005). Converging evidence: Bringing

together experimental and corpus data on the association of verbs and constructions. Cognitive

Linguistics 16(4): 635-676.

Gries, Stefan Th., Beate Hampe & Doris Schönefeld (to appear). Converging evidence II: More on

the association of verbs and constructions. In Newman, John & Sally Rice (eds.) Empirical and

Experimental Methods in Cognitive/Functional Research. Stanford, CA: CSLI.

Gries, Stefan Th. & Anatol Stefanowitsch (2004a). Extending collostructional analysis. A corpus-

based perspective on ‘alternations’. International Journal of Corpus Linguistics 9(1): 97-129.

Gries, Stefan Th. & Anatol Stefanowitsch (2004b). Covarying collexemes in the into-causative. In

Achard, Michel & Suzanne Kemmer (eds.) Language, Culture, and Mind. Stanford, CA: CSLI.

225-236.

Gries, Stefan Th. & Anatol Stefanowitsch (to appear). Cluster analysis and the identification of

collexeme classes. In Newman, John & Sally Rice (eds.) Empirical and Experimental Methods

in Cognitive/Functional Research. Stanford, CA: CSLI.

Hollmann, Willem (2003). Synchrony and Diachrony of English Periphrastic Causatives: A

Cognitive Perspective. Unpublished PhD Thesis. Manchester: School of English and

Linguistics, University of Manchester.



45

Kastovsky, Dieter (1973). Causatives. Foundations of Language: International Journal of

Language and Philosophy 10: 255-315.

Kemmer, Suzanne (2001). Causative constructions and cognitive models: the English ‘make’

causative. Paper presented at the First Seoul International Conference on Discourse and

Cognitive Linguistics: Perspectives for the 21st Century. Seoul, 9-10 June, 2001.

Kemmer, Suzanne & Arie Verhagen (1994). The grammar of causatives and the conceptual

structure of events. Cognitive Linguistics 5(2): 115-156.

Kirchner, Gustav (1952). Die Zehn Hauptverben des Englischen im Britischen und

Amerikanischen. Halle (Saale): Max Niemeyer Verlag.

Langacker, Ronald W. (1991). Foundations of Cognitive Grammar. Volume 2. Descriptive

Application. Stanford, CA: Stanford University Press.

Levin, Beth (1993). English Verb Classes and Alternations. A Preliminary Investigation.

Chicago/London: The University of Chicago Press.

Murphy, Raymond (1985). English Grammar in Use. A Self-study Reference and Practice Book

for Intermediate Students. Cambridge: Cambridge University Press.

Newman, John & Sally A. Rice (2006). Transitivity schemas of English EAT and DRINK in the

BNC. In Gries, Stefan Th. & Anatol Stefanowitsch (eds.) Corpora in Cognitive Linguistics:

Corpus-Based Approaches to Syntax and Lexis. Berlin/New York: Mouton de Gruyter. 225-

260.

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik (1985). A Comprehensive

Grammar of the English Language. London/New York: Longman.

Siewierska, Anna (1991). Functional Grammar. London/New York: Routledge.

Sinclair, John M. (1991). Corpus, Concordance, Collocation. Oxford: Oxford University Press.

Stefanowitsch, Anatol (2001). Constructing Causation: A Construction Grammar Approach to

Analytic Causatives. Unpublished PhD thesis. Rice University, Houston, TX.

GAËTANELLE GILQUIN


46

Stefanowitsch, Anatol & Stefan Th. Gries (2003). Collostructions: investigating the interaction of

words and constructions. International Journal of Corpus Linguistics 8(2): 209-243.

Stocker, Markus (1990). The Causative in Middle and Modern English. Unpublished PhD Thesis.

Muttenz: Universität Basel.

Stubbs, Michael (1995a). Corpus evidence for norms of lexical collocation. In Cook, Guy &

Barbara Seidlhofer (eds.) Principle and Practice in Applied Linguistics. London: Oxford

University Press. 245-256.

Stubbs, Michael (1995b). Collocations and semantic profiles. On the cause of the trouble with

quantitative studies. Functions of Language 2(1): 23-55.

Thomson, Audrey J. & Agnes V. Martinet (1980). A Practical English Grammar. Third Edition.

Oxford: Oxford University Press.

Van Ek, Jan A. (1966). Four Complementary Structures in Predication in Contemporary British

English. An Inventory. Groningen: J.B. Wolters.

Van Ek, Jan A. & Nico J. Robat (1984). The Student’s Grammar of English. Glasgow: Bell &

Bain Ltd.

Van Roey, Jacques (1982). English Grammar. Paris: Didier Hatier.

Visser, Frederikus Th. (1973). An Historical Syntax of the English Language. Part III, Second

half: Syntactical Units with Two and with More Verbs. Leiden: E.J. Brill.

Werner, Patricia K., John P. Nelson & Lida R. Baker (1990). Mosaic II. A Content-based

Grammar. Second Edition. New York: McGraw-Hill Publishing Company.

Wierzbicka, Anna (1998). The semantics of English causative constructions in a universal-

typological perspective. In Tomasello, Michael (ed.) The New Psychology of Language.

Cognitive and Functional Approaches to Language Structure Mahwah, NJ/London: Lawrence

Erlbaum Associates Publishers. 113-153.

The verb slot in causative constructions. Finding the best ... · Constructions SV1-3/2006 (,...

Documents

Transcript of The verb slot in causative constructions. Finding the best ... · Constructions SV1-3/2006 (,...